Hands-On Industrial Internet of Things

Data-driven

Data-driven methods can use ML, statistics, or Artificial Intelligence (AI), deep learning. These techniques depend on collecting a history of failures, which requires volumes of data.

Without having a comprehensive understanding of the system, it can be hard to know how much data is good enough for a specific purpose. We suggest that you collect at least six months worth of data with relevant events. In data-driven approaches, the most common technique is to use an artificial neural network (otherwise known simply as an NN or deep learning, in which a network model learns a way to produce a desired output. In the CBM, for instance, this might refer to the level of degradation of a turbine or the lifespan of a filter.

Another ...